Sponsored by: National Science Foundation
Research Applied to National Needs (RANN)
Division of Advanced Environmental Research
and Technology
Earthquake Engineering Program, Grant GI 39644
ee
FINITE ELEMENT ANALYSIS OF A SEISMICALLY
EXCITED CYLINDRICAL STORAGE TANK,
GROUND SUPPORTED, AND PARTIALLY
FILLED WITH LIQUID
Samia Hafiz Shaaban
William A. Nash
Department of Civil Engineering
University of Massachusetts
Amherst, Massachusetts 01002
duly, 1976TABLE OF CONTENTS
ABSTRACT
BACKGROUND
ANALYSIS
Governing Equations
Boundary Conditions
Finite Element Ideali zation
Free Vibrations of the Coupled System
Response of the Coupled System to Base Excitation
Modal Analysis Solutions
Reactions of the Base
Liquid Oscillations in a Rigid Container
COMPUTER IMPLEMENTATION
Computer Prograns
Nature and Size of System Matrices
EXAMPLES.
Free Vibrations of Completely Filled Rigid Tank
Free Vibrations of a Completely Filled Elastic Tank
Free Vibrations of a Partially Filled Elastic Tank
Cylindrical Tank Whose Base Slab is Subject to
Artificial Earthquake Excitation
ABRIDGED METHOD OF COMPUTATION
. ‘ACKNOWLEDGMENT
REFERENCES
APPENDIX A. DEVELOPMENT OF MATRICES EMPLOYED IN FINITE
ELEMENT ANALYSIS
APPENDIX B. PROGRAM LISTING FREE VIBRATIONS OF LIQUID
FILLED RIGID TANK
” APPENDIX C. PROGRAM LISTING - FREE VIBRATIONS OF LIQUID
FILLED ELASTIC TANK
APPENDIX D. PROGRAM LISTING - RESPONSE OF LIQUID FILLED
TANK TO BASE EXCITATION
4
44
57
59
67
69
70
Ad
B-1
D-1ABSTRACT
The structure under consideration is an elastic cylindrical
liquid storage tank attached to a rigid base slab. The tank is filled
to an arbitrary depth with an inviscid, imcompressible liquid. A
finite element analysis is presented for the free vibrations of the
coupled system permitting determination of natural frequencies and
associated mode shapes. The response of the partially-filled tank to
artificial earthquake excitation is also determined through use of
finite elements. Examples, together with program listing, are offered.uu)
BACKGROUND
A previous report [1] by these same investigators developed a
Finite element approach for determination of small amplitude elastic
responses of an empty slab-supported cylindrical liquid storage tank
subject to arbitrary base excitation. It was assumed that the base
slab supporting the tank is rigid and that the tank does not separate
from the slab during excitation. The present investigation continues
the work presented in [1], but with the significant addition of an
inviscid, incompressible liquid filling the tank to an arbitrary depth.
Again, finite elements are employed to represent both the elastic tank
as well as the liquid. Natural frequencies and associated mode shapes
of the coupled liquid-elastic system are found through use of finite
elements. Also, the special case of the natural frequencies and associated
mode shapes of a liquid in a rigid container is investigated. Next,
using modal superposition, a program is developed for determination of
the response of the coupled liquid-elastic system to arbitrary base
exci tation.(2]
ANALYSIS,
Governing Equations
For the elastic circular cylindrical tank with a vertical geo-
metric axis under consideration here, we shall employ a series of
ring-shaped finite elements extending from the base slab-to the tank
top, with each ring being bounded by a horizontal plane normal to the
shell axis. Both in-plane as well as out-of-plane displacements and
forces in the shell must be considered. Again, as in [1], the shell
theory due to J. L. Sander, Jr. [2] is employed to represent the small,
elastic deformations of the cylindrical tank. Let the radius of the
tank be R and its thickness beh. Further, let the quantities r, 8,
and z denote radial, circumferential, and axial coordinates respectively
of a point on the middle surface of the shell. The corresponding dis-
placement components are denoted by w, v, and u. The equations of
motion of the elastic tank in terms of w, v, and u are given in [1].
The liquid in the tank is assumed to be homogeneous, incompressible,
and inviscid. Further, the flow is taken to be irrotational and only
smaii amplitude liquid motions are considered. Lastly, it is assumed
that there are no Sources, sinks, or cavities anywhere in the liquid.
Under these conditions the motion obeys the Laplace equation
vp (r, 6, 2) = 0 a)
where p represents total pressure at any point. The total pressure
is the sum of the static and dynamic pressures, viz:[3]
P= Pst * Payn
where p,, is the pressure that would exist if there were no motion
and Pyyq arises because of motion of the liquid. Since the static
pressure obeys Laplace's equation, obviously the dynamic pressure
does also. Henceforth, the dynamic pressure will be denoted by p
for brevity.
The Bernoulli equation may be expressed in the form:
gz + Bo Pts yes Bo (2)
Pe PE ot
where z is as defined for the shell with origin at the \iquid surface,
g is the gravitational constant, pg denotes liquid density, v the
magnitude of velocity at any point in the liquid, t denotes time,
and ® is the velocity potential. Since the liquid is nonviscous, the
motion is irrotational, and the oscillations are of small amplitude,
the velocity squared term in (2) may be neglected in comparison with
other terms. Also, for z measured positive upward from the liquid
surface we have:
P,
ae 0 (3)
Thus, (2) becomes:
Py Be.
a taf 0 4)(4)
Boundary Conditions
At the liquid free surface, the vertical velocity component is
given by:
vy (rs 85 0) = | free) [2] (3)
a
20
1
9
where € is the superelevation of the free surface over the undisturbed
surface level. The linearized free surface condition may be expressed
in the form:
2,
ao ao
odie O ogy |e 0) (6)
[sox]
Using (4) together with the relation o-gé = p, this may be expressed
in the form:
2,
1 p,m.
Ge 7
For the liquid under consideration the velocity vector V may be
written in the form:
V = grad
ee
(8)
Consequently, the boundary conditions expressing liquid-solid inter-
action along the elastic wall of the cylindrical tank as well as at
the rigid bottom of the tank may be written as:(5)
ow
[| in the wetted part of the tank wall
cone r=Rk
(3)
[o] at the rigid tank bottom
ae
Here, 1 is a unit vector normal to the liquid-shell boundary and
H denotes depth of liquid in the tank. Thus, along the wetted
elastic tank wall denoted by = we have:
ao Ow a
[2 - =| :
FR
where w is the radial displacement of the tank wall at any point
(R,z,6). Again, using (4), this becomes:
2
4.22 7 (10)
[% Pe ar |
FR
Since the liquid velocity in the z-direction is zero at the tank
bottom, it follows from that:
a)
In summary, motion of the liquid is completely defined by the
Laplace equation (1) together with boundary conditions.(6]
In a finite element approach to the coupled liquid-elastic tank
problem, the finite element matrix equation is obtained either from
the governing differential equation by using Galerkin's method, or
from the variational equation by using a minimization technique [3].
Use of the Galerkin procedure necessitates knowledge of the governing
differentia] equations of motion together with selection of a weighting
function which may be chosen to be the same as the element shape
function. Setting the first variation of the resulting integral equal
to zero yields the desired finite element matrix equation. Use of
the Euler-Lagrange method necessitates formulation of the kinetic
energy (found by integrating over the liquid voTume), the potential
energy (found by integrating over the free surface), and the work done
on the liquid by external effects (such as solid-liquid interface
forces). Minimization of energy then yields the governing equations.
In [3], it is demonstrated that both approaches yield the same finite
element matrix equation provided the same type of element and the same
shape function are employed in both treatments.
In [3], it is shown that an appropriate variational functional
for the liquid is
t,
ref? r= n- wat (1a)
4
where T, Il, and W represent the kinetic energy, the potential energy
of the liquid, and the work done on the liquid respectively. These07]
a are given by [3]
: T =(12) 05 { vo + vodv
’
a=(/2)[ Elo, 96) 6s
(13)
F
We Pf ay ods
z
where p¢ denotes liquid density and £ is the deviation of the liquid
elevation from the static configuration. The kinetic energy is evaluated
by integration over the liquid volume V, the potential energy by inte-
gration over the free surface F, and the work by integration over the
Tiquid-tank interface 2.
In the present investigation, it is most convenient to investigate
the dynamic problem in terms of the liquid dynamic pressure p. If
damping is neglected, this leads to a matrix differential equation
involving only the pressure together with its second derivative with
respect to time. In [3], Eq. 3.9 it is shown that the functional per-
tinent to the governing equation (1) together with boundary conditions
(7) and (10) may be written in the form:
2
1/2)/ Ws vp dy =k [c2y?as = 04 [ep Was
4 : i at’
(4)
where the definitions of I), I,, and I, are evident from (14).[8]
Finite Element Idealization
The liquid is discretized into annular elements of rectangular
cross-section. These elements may by considered to be formed from
the intersection of concentric annular cylindrical surfaces with a set
of horizontal planes. The intersection of these surfaces with, the
planes gives rise to nodal circles, as shown in Figure 1.
@0
FIGURE 1
This three-dimensional problem can essentially be transformed
into a two dimensional one by developing the pressure p in a Fourier
series in the circumferential direction, viz:
par hy cos mo (15)
The problem of forced motion of the slab supported tank when excited
by horizontal ground accelerations can be reasonably well described
through consideration of only the first harmonic, m= 1 provided that(9]
one is concerned with obtaining the motions about the neutral equili-
brium configuration. However, for the sake of generality, the following
finite element matrices will be developed for an arbitrary number of
harmonics min the circumferential direction. Thus, let us set
Py (fs 2, 6) = Ph (r, z, 0) cos mo (16)
Henceforth, the subscript m will be omitted for brevity.
Thus, the problem has been reduced to a two dimensional one in
the plane indicated by r, z, 0 = 0 in Figure 1. Henceforth, we shall
use (x,y) as local coordinates, which origin at the geometric center
of the element, to denote the position of any point in this plane. The
liquid pressure at any point in this plane is described using the nodal
Pressure parameters of the corresponding rectangular element surrounding
it. Thus:
Plxsy) = EN} 4&)) (7)
where [N] represents the element shape function and {6,) is the element
nodal pressure vector. The shape function is obtained by assuming a
suitable interpolation function which here is taken to be a linear
variation of liquid pressure in both the x and y directions. Thus:
Py
5p
6P3
504
Pxay) = gg E(a-x)(b-y) (24x) (b-y) (a4x)(b4y) (ax) (D4y)
(18){10}
Figure 2 indicates a typical element of length 2a in the radial (r)
direction, 2b in the z (axial) direction, whose center (0,0) lies
at a distance x, from the geometric axis of the tank.
FIGURE 2
Liquid Element
From (16), we have:
vp = 2 cosima)7,, + # cos (ne), - ® sin(me) PF,
2
wp vp 2? cos” (me) + (2)? cos? (mo) + 85 sin? (me) P2
4
(19)
(20)oy
It is now necessary to determine the functional (13). Substi-
tution of the pressure (15) into the integral defining I, and inte-
gration over the liquid volume v yields:
(1/2)| vp + vp dv
2
-072){ { { (a? cos 46) + ey? cos4na)+ My singe] rdedz dr
rz6 Fe
a
5 ((e +e? “5 dr dz
ra
2 p2
= BL { |B? rey? 4 P| (xy + axe
i | ae (x +) oo (21)
oe
(1/2) 16,)"[K,1 (6) (22)
The element stiffness matrix [K,] is developed in detail in Appendix A.
The integral defining I is found by integrating over the liquid free
surface F to be:
toed [eras
2°29 IGE
Fne}
)? (xy + x) dx
a
F [6 PTO INI Gp) (ag + 0) ae (23)
x
= 1/2 6)" IM] 6) (24)
(es DA a)
The element mass matrix [M,] is found using (18) and is given in detail
‘in Appendix A. The integral defining I, is found by integrating over
the liquid-elastic shell interface = to be
2
= a] S
13 = oF |p a a.
z e
= bg { [P ay cos” (ne) R dodz
at
8
2, :
= pg fp Wa (25)
at’
2
where R is the tank radius and
w(z, 6) = W(z, 5) cos (may (26)
The generalized radial displacement of the tank W may be represented
in terms of the finite element generalized coordinates {6} through
the following:
W(z, 0) = [Ny] {5} (27)13]
Thus,
(28)
(29)
From this the force matrix (s,] representing the coupling effect is
determined. This is developed in detail in Appendix A. The assenbled
liquid mass and stiffness matrices are denoted by [Mp] and [Kg]
respectively, and the coupling force matrix is assembled in [S].
The partial differentia) equations, in matrix form, governing
liquid motion may be found by first realizing that the functional
1 (14) is of the form:
t.
2 . .
= A eee eee a 30)
re [4 1° Spa" 01° 8p, ) at (30)
a
Then, setting the first variation of this equal to zero, viz:
61-0 au
An Euler-Lagrange equation for each independent variable 6,
i
may be obtained from the expression:
(32)4]
Substitution of (22), (24), and (29) into (14) yields:
T= 1/2 (6) Ekp](5 3 - 1/2 (5) EMp]S,) - petd,)ESIS} (33)
eee | dah ine oe u
Thus, (32) leads to:
Ukp]&6,} + [Mp](5,} - ofS]és,3 = (OF (34)
Also, the equation of motion of the elastic shell may be written in
the form:
(M}t5) + (K]6y} = (6p) (35)
where {6-} denotes the generalized force vector at (z, 0) which may
be expressed as
{5p} = {5-3 + {6_ } (36)
F FS Fp
where (5, } represents external nodal forces including the static
e
pressure of the liquid and {6- } represents nodal forces exerted on
P
the shell arising from oscillations of the liquid. Also, (MJ and [K]
are the shell mass and stiffness matrices corresponding to a prescribed
circumferential harmonic nunber m.
Free Vibrations of the Coupled System
Since we are interested in the free vibrations of the shell
about the static equilibrium configuration (35) yields:
IMIG) + EIS, = Cp (37)a5]
The generalized force vector corresponding to the dynamic
pressure p,, on the inner surface of the shell is given by [5]:
Pn
Gp) == aR [INIT ONI(9,) dz
p =
et
= - [s]'6}
. : 7
“ [MI{5,} + [k]{5,} + [5] {55} = 0 (38)
Thus, the free vibrations of the coupled liquid-elastic tank
system may be expressed in the form:
: ~P¢S Mp is oO ke
: Pp
at (39)
Let us redefine the mass and stiffness matrices of the liquid as:
(40)
| 9 (4)fe)
These system matrices are nonsymmetric and extraction of eigen-
values and modes becomes extremely difficuJt, particularly when very
large size matrices are concerned. In view of these serious numerical
difficulties, let us adopt the approximation suggested by Hsiung and
Weingarten [3] which consists of neglecting the free surface boundary
condition (5). This simplification implies that (a) the liquid mass
matrix [M¢] corresponding to the free surface potential energy vani-
shes, and (b) the free surface pressure is zero, It is to be noted
that in the present investigation the free surface condition was
evaluated at the mean liquid level. Thus, the degrees of freedom
corresponding to the free surface are constrained and can be omitted.
Because of (a), we immediately have:
~ [8] 6} + Ukp]€6,} = 0
. Peo (42)
(6p) = Kel SH)
Thus:
.
(Mytsy) + (kts, + (81745,) = (0)
(43)
Lon + sisi] 6) + HIG) = 0
7 wu wu
This means that the shell mass matrix is augmented by an added mass
matrix:
from) = (s1"tk,J"'¢s7 (44)07
For the case of free vibrations of the system
6) = -u? (6,) (45)
u 0
where w is the natural frequency of the coupled system and the
equation for eigenvalues is:
| [Sut
——} +
My " 6
ut
which #s identical with Equation (2) in [1]. Here, {5,4} and {8,41
are the known support displacements and accelerations, respectively, andO91
[Syp) and (5,4) are the total off-base displacements and accelerations
corresponding to this response analysis.
All elements in the top line of Equation (49) pertain to base
node parameters. Thus, Ky, and M,., denote forces at base nodes due
to unit displacements at the base nodes and the superscript T, of
course, denotes matrix transpose. ky and M, in the bottom row are
coupling effects between the base nodes and the other (non-base)
nodes. All other elements in the bottom row of Equation (49) pertain
to non-base nodal parameters. Thus, K and M are’ redefined to represent
stiffness and mass matrices of all non-base nodes.
At any time, the displacement vectors of the non-base nodes can
be considered as a summation of two vectors. The first vector {U,}
is a function of the instantaneous ground displacement, thus it can
be called static. The second vector (U4) is a function of the ground
acceleration history, thus it is termed dynamic.
This approach furnishes a suitable method to reduce the equations
of motion to the familiar form of forced vibrations:
IM](Ug) + [K]Ug} = (FD (50)
} = (U} + Wy) (51)[20]
The equations of motion are:
7 : .
é aa 6
Ie WL Abe |e | & ( we | |p (52)
iM " feo, oF (4 |
The equations of the off-base elements are
TH] yye) + (MICU,) + EMU? + Ok Heyyy? +
(53)
[k]{U,} + [k]{Uy} = 0
Now it is attractive to define U, as a displacement vector so
that when it is associated with the ground displacement vector U, +
the resulting motion of the structure corresponds to no internal
strain energy. Hereafter, 54); will be denoted by U,, for brevity.
This condition implies that:
(K,JtU,,) + [K1tU,) = 0 (54)
In other words, the vector {U,} is developed through rigid body
displacements consistent with {U,,}. Thus, from (54)
= uy!
(Ug) = -UKT EK) Ope
This phenomena has also been demonstrated numerically and the
resulting static displacement U, is nothing but a series of U,, or[a1]
Ys) oe
Use Yoe
U, U
tu.) = i (55)
Yon Ube
where N is the total number of elements and {U.;} is the displacement
vector of node i = (Uy,) for all values of {and (Uy) is a (4 x 1)
vector representing the axial, tangential, and radial displacements
as well as the rotation of the generator at the base.
Thus, the off-base node equations yield
IN](Ug) + CKIUg) = -[¥,]U,,) - [MICU
(MUU) + OKLUg) = -C0M] - CMIOKT EK, 1,
= [effective mass matrix] (U,4)
= Meee] Ue? (56)[22]
It should be pointed out that for most practical tank dimensions
the- driving forces developed due to the mass [M][K][K,] are much
larger than those developed by [M,]. This has been demonstrated
numerically.
The ground acceleration vector U,, will be proved to be equal to:
0
TD 1
Ug(t }
where Ug(t) is the ground acceleration amplitude at tine t.
Since the base of the tank is excited by a ground displacement
and acceleration acting in its plane and in the constant direction
8 = 0, no axial acceleration component develops and the ground accel-
eration will be completely defined by its amplitude value Uytt):
Uglt) = Peak + #(+) 7)
The peak is an acceleration value independent of time and f(t) is a
non-dimensional function of time.
The associated base-node displacement vector U., is derived by
use of Fig. 3, viz:
u(o,6,t) = 0
v(0,8,t) =-Ug(t) + sin @ = -Peak + f(t) + sino
w(0,8,t) = -Ug(t) + cos @ = +eak + f(t) + cos 6
Mo,0,t) = 0.0 (58)[23]
FIGURE 3
Since the excitation function is described in the previous form
to be associated with m= 1, obviously only the first circumferential
harmonic will be excited, and thus the vibration of the tank can be
Prescribed by superposition of certain contributions of different.
axial modes corresponding to m= 1 only (see Appendix A, in [1], for
assumed form of loads and displacements) .
.
Up(t) = Peak + #(t) + |“
a
(39)[24]
Let
0
: lel
Were) cee rd a)
0
., The equations of motion reduce to:
IM]{Ug} + [K1(Ug} = (Poggh + Ft)
which is the desired form of forced vibration to which the modal analysis
technique will be applied.
Modal Analysis, Solutions
[M]Ug) + [K]{Ug} = {Pope > f(t)
Let
fy) = DA®
{Ug} = DXA}
., [X] is the rectangular mode matrix formed as a set of
mode vectors (n x k) where
n = number of degrees of freedom of the non-base elements
k = number of modes considered in the analysis
{A}= mode participation factor vector = k x 1
J. [MIDIAY + OKIEKHAD = Pyggh + Ft)
{Ay = (A(t) 5D = A(t}
U,(t)
f(t) = Be[25]
Premul tiply by (X1"s (k x n)
. T, " T Be iT es
w DTIMIDCAY + [XD TKIEXI(AY = [XT Poggh + F(t) =
fer} + F(t)
Now, use the orthogonality condition:
OTK = 0 ken
Obviously the resulting matrix [X]'[MI[X] = [6M] is a diagonal matrix
since the (generalized k x k mass matrix) nonvanishing terms are only
Dx," EMIDX,] = GM(nsn).
The same concept holds for [X]"[KI[X] = [6S]
diagonal matrix
where 2 is the squared eigenvalue diagonal matrix = [0°][6M):
a
1
kKxk
Thus, GM, as well as GS can be considered as vectors,
eB] 2
GM(1,1) GM) wt
GM(2,2) j
and respectively.
6M(3,3) 6m(2.2) w2
GM(k sk) pe °2
Gw(k.k) wo?
n[26]
Thus, k independent equations result:
GM(I,1) + ACI) + a(1) + w(1) + ACL) + GM(I,T) = @P(I) + #(t)
where I refers to the mode number .
Ray + wc) + acy = thy « ete)
which are the equations of k independent lumped masses each representing
the participation of the corresponding I-th mode,
Now, A(I) can be found using Duhamel integration to account for
the initial conditions (just before the instant t), i.e. to consider
the whole acceleration record imposed on the structure, viz:
te
ACI) = GPCL : { f(x) + sin u(ter)dr
0
t
where PIN(I) = ( f(r) sin a{t-t}dt) + GP(1)
°
J AUD = felts Fey = Got? + ac)
Now from the origina? equations of motion the displacement and
acceleration nodal vectors are determined:
(U4) > DH)
tp = Day
The accuracy of the modal analysis approach depends on the number
of modes involved in the superposition. The latter depends on how close
or scarce the natural frequencies of the structure are spaced.(27)
The accuracy of the method can be examined through the satis-
faction of the original external equilibrium equation:
TH] (Ug) + [K] Ug) = (agg) + Ft)
For the structure considered, it was found that the superposition
of a few modes offered only a crude approximation since the external
equilibrium equation failed to be satisfied by as much as thirty percent.
Use of ten modes reduced the maximum discrepancy to about ten percent.
Reactions of the Base
From the equations of base vibrations:
U
T bt T
TH, 145] eat * Coble?
s + Ug
Now, {U,} and {U,} were proved to be equal to:
1 I
1 I
fu} = 1 r «
{Upp} and (U.} = Upset
I 1[28]
where [I] is a (4x4) identity matrix, N/4 of which form the relating
matrix between the resulting static non-base node displacements and
‘the base node imposed displacements. Also N = number of non-base
node degrees of freedom and since " contains nonzero elements only
in the first four colums Mj. U, can be expressed as:
IME} C1104?
b be
where [M[]' is the 4x 4 matrix including the nonzero elements
My + ME TDU yg) + EMT CU) + ky + KE'THCU,)? 0K, Uy) = CF}
Ty
but (Ky, + Ky‘ 1U, 4)
: : host 1
2's GpF = yy + MTD, + IMU) + [xu (60)
Of course, the most significant part of the base force is attributed
to the displacements of the non-base nodes, i.e. [k,]"(U,).[29]
Liquid Oscillations in a Rigid Cylindrical Container
The fluid dynamic pressure arising due to liquid motion in a
rigid cylindrical tank will be governed by a special case of (34).
Since the rigid container degrees of freedom {6,} are restricted,
{6,} can obviously be omitted and the governing equations yield:
Ukg]C6,) + [Mg](6,) = (0 (61)
Although the fluid "mass" matrix [Mp] is defined to be Nope x
Nope (where Nore is the total nunber of degrees of freedom of the
liquid), the nonzero elements are those corresponding to the free
surface generalized pressure vector only. A matrix condensation
approach is employed to minimize the computer Storage area as follows:
th 2 |} (62)
where the second set of equations corresponds to the free surface
nodes (n, in number) and the first set corresponds to the remainder
of the liquid nodes (nj). This leads directly to:
1 8, * Ki2 8p 79
(63)
-1
8, = nT! Kaley)[30]
Substituting this into the second set of equations (61) yields:
T -1 =
Chie ii Sig ate cpes ee o (64)
[Keonal 6) + [Meonal 16, =0 (65)
where
Bae T yo
Ueonad = [Koa] - Kya] (Ky, J 'tKy2I
The condensed stiffness matrix
Myo = (mp x ng) (67)
(my Xn,) (66)
Meonad
= The condensed mass matrix
The submatrix Ky» (nj x np) also has a significantly smaller nonzero
submatrix =n x Ny and the second matrix of (63) can be efficiently
evaluated by use of this fact as follows:
“1
0
n
Ky 1
1 Wat
Kiel
—_— 1
The direct inversion of [ky] is avoided and the last (ny x ny)
matrix resulting from the multiplication ky,"'Ky2 15 the only portion
treated, through the use of Gaussian elimination back substitution [4].
This, in fact, corresponds to the generalized nodal pressure vector
{8pq} of the row immediately below the free surface.
Therefore, in the assembly of the stiffness matrix three sub-
matrices are considered: Kj, = my x M3 Kyo = ny X My (non-zero terms[31]
are ny x np), and Koy = My x My. tn the assembly of the mass matrix
only the Mp9 = ny x Ny matrix is formulated.
The liquid matrices numbering schemes (for a rigid tank) are
given in Figures 4a through 4d.
FIGURE 4a
Liquid Degrees of Freedom numbering scheme pertinent to
the stiffness matrix generated in program RIGID for
symmetric harmonic modes.
¢
MMC HNN) #1 [WL (M1) (NNT)
rt ma(NN+1)
wall
2(NN)+3 Pt ; tf 3(NN#1)
nme? | 2(nns1)
Rese eemee
J=1, MMT t 1 2 3 ieee NN+T ‘ NNNH)
* mean water level[32]
FIGURE 4b
Liquid Degrees of Freedom numbering scheme pertinent to the
condensed mass matrix generated in program RIGID for synmetric
harmonic modes.
NNeT
T (+1 NN
MM(NN)
fe
2NN + al
3NN
eT ,MMe1 | 2NN
Peeeeeeeeeeee tee ee
eee ae
NN +1 ——> 1 = 1,NN
FIGURE 4c
Liquid Degrees of Freedom numbering scheme pertinent to the
stiffness matrix generated in program RIGID for asymmetric
harmonic modes.[33]
COM = (NNN)
FIGURE 4d
Liquid Degrees of Freedom numbering scheme pertinent to
the condensed mass matrix generated in program RIGID for
asymmetric harmonic modes.[34]
COMPUTER IMPLEMENTATION
Computer Programs
Three separate programs were developed in the present work.
The first, program RIGID, determines liquid oscillation natura
frequencies and associated mode shapes in a rigid circular cylin-
drical container fixed to a rigid base. In the early stages of this
work this program served as a check on the formulation of the liquid
mass" and "stiffness" matrices and thus on the validity of the entire
liquid idealization process. This is because in many cases, the
data obtained were in good agreement with existing work involving
rigid containers.
The second program, COUPLE, is employed to investigate natura’
frequencies and associated mode shapes of the coupled liquid-elastic
tank system described by Equation (46). To this end the first main
program described in [1] (MAIN) was modified slightly so as to cor-
respond to two sets of ring-shaped finite elements representing the
cylindrical tank. The first set of elements corresponds to the lower
(wetted) surface of the tank and the second set to the portion of the
tank above the liquid level (dry). The program corresponding to this
representation is henceforth termed SHELL. A single run string was
prepared of COUPLE and SHELL so as to be able to investigate the
coupled liquid-elastic tank system. This also serves to retrieve
the “added mass matrix" stored on a disc file by program COUPLE and
to then add its terms to the corresponding shell mass matrix terms.[35]
Program COUPLE carries out the following operations: a) It
devises a numbering scheme for the liquid finite element mesh. This
is accomplished in subroutine FLGEN which requires as input the
nunber of liquid finite elements NN along the tank radius in a single
row, the number of liquid finite elements MM in a single colum,
and the specified number of circumferential harmonics, m. This is
illustrated in Figures 5a and 5b.
FIGURE 5a
Liquid Degrees of Freedom nunbering schene generated in
program COUPLE for asymmetric harmonics pertinent to the
liquid "stiffness" matrix. (m= 1, 3, 5, ....)
t
MeWeL.* =
™ 3M 2
cfc ee 2
5
°
=
=
=
g
=
: (NN=1) :
hc me |S
3
TPMT MIT (NN-1) 4
—F1=1,NN
adius of = IN] elements )
* mean water levela
[36]
¢
Lape Re
mt MH) (NNT)
1 MM
. =
4 8
' =
. ‘s
1M 2 5
1 8
TAT NN (MM)
Radius of tank
I = 1,NN+1
FIGURE 5b
Liquid Degrees of Freedom numbering scheme generated in
program COUPLE for synmetric harmonics pertinent to the
liquid "stiffness" matrix (m= 0, 2, 4, ....)
b) It evaluates a set of different liquid stiffness and coupling
element matrices [K,] and (s.1, each corresponding to a column: of
elements in the liquid idealization scheme. It is assumed that the
Tiquid has been discretized into equal rectangular areas. This is
accomplished in subroutines FSTIFF and FFORCE. c) It assembles the
liquid stiffness matrix [kK] im accordance with the numbering scheme
mentioned in (a) above into a half-banded matrix stored in a linear
array so as to minimize core allocation. The condensed coupling matrix
is also assembled into an (MM, 2MM) matrix, [S].d) It evaluates the
Viquid added mass matrix defined in (44) and stores it on a disc file
to be retrieved by SHELL.[37]
The third program, RESPONSE, which follows after SHELL in
the run string, accomplishes the following:
e) It evaluates the generalized forces developed at the tank wall
nodes due to a unit ground acceleration in the horizontal direction.
f) It transforms the system properties into modal coordinates. That
is, the generalized mass vector GM and the generalized force vector
GP are evaluated. These operations are performed in the first
section, PARTI. g) It retrieves in PARTII the ground acceleration
record ACC previously generated utilizing program PSEQGN available
through the National Information Service-Earthquake Engineering -
Computer Program Applications, and which was stored on a disc fite
[7]. To improve the accuracy of the response computation the total
time history under consideration is arbitrarily divided into smaller
time intervals by "guiding" time stations, the modal velocities {A}
and displacements {A} of which are first determined independently in
subroutine CONTROL. CONTROL calls subroutine RES at each time station
to evaluate the Duhamel integral of the previous acceleration record.
The vectors {A} and {A} are stored in the core array to be used as
illustrated below:
h) It evaluates the specified nodes generalized displacements and
prints the response history and stores it in disc files to be re-
trieved for automatic plotting purposes. The responses of the speci-
fied degrees-of-freedom designated as NDI, ND2, and ND3 are stored on
tapes number 4, 5, and 6 respectively. These degrees-of-freedom are
explained in detail on page (60) together with Figures 17 and 19.[38]
Knowing the response history at any degree-of-freedom, the corresponding
stresses can be found from the program RESP given in [1]. This is with
regard to the internal forces developed, the reactions at the tank base,
and the force equilibrium check if so desired.
Nature and Size of System Matrices
The original sizes of the system matrices are indicated in
Equation (46) together with the numbering schemes shown in Figures
4 and 5 to be indicated below. For brevity, the following programming
symbols were employed:
8, = NDFST x1
p 7 NOFF x1
Mand K = NDFST x NDFST
Mp and Kp = NDFF x NDFF
S = NDFF x NDFST
where
NN = number of liquid element in one row along the
tank radius
"M = number of liquid elements in one column along the
tank generator
MMT = total number of shell ring elenents
NDFST = total number of shell degrees of freedom
= A(MMT + 1)
NDFF = I.d
where
I = NN for asymmetric harmonic modes
= (NN +1) symmetric harmonic modes
J = MM for the coupled case with zero pressure assumption
at the free surface
= (MM + 1) for the fluid oscillation in a rigid cylin-
drical container